Finding the Spatial Co-Variation of Brain Deformation With Principal Component Analysis

Area of Science:

• Biomechanics of traumatic brain injury (TBI)
• Computational modeling and simulation
• Machine learning applications in injury prediction

Background:

• Traumatic brain injury (TBI) assessment relies on metrics like strain and strain rate.
• Traditional finite element (FE) head models use numerous elements, leading to complex spatial data.
• Brain inertia influences injury metric distribution, suggesting potential for more concise pattern representation.

Purpose of the Study:

• To apply principal component analysis (PCA) to identify spatial co-variation patterns of TBI injury metrics.
• To investigate these patterns across different head impact types (simulation, football, MMA, car crashes).
• To improve machine learning head models (MLHM) using PCA-derived injury metric patterns.

Main Methods:

• Principal component analysis (PCA) was used to decompose injury metrics (maximum principal strain, MPS rate, MPS × MPSR) for various impact types.
• The first principal component (PC1) was analyzed to understand spatial co-variation.
• A machine learning head model (MLHM) was developed to predict PC1, enabling inverse transformation for all brain elements.

Main Results:

• PC1 explained significant variance across all analyzed datasets.
• The corpus callosum and midbrain showed high variance in injury metrics across all impact types.
• The PCA-enhanced MLHM achieved a 74% reduction in model parameters with comparable maximum principal strain (MPS) estimation accuracy.

Conclusions:

• Brain injury metrics can be effectively decomposed into mean and PC1 components, capturing high explained variance.
• Spatial co-variation analysis provides enhanced interpretation of injury metric patterns.
• PCA significantly improves the efficiency of machine learning head models for TBI analysis.